Research Article
Obfuscated Tor Traffic Identification Based on Sliding Window
Table 1
Analysis of nonobfuscated Tor traffic identification techniques.
| Reference | Feature selection | Type of algorithm/method | Dataset | Evaluation metrics |
| Lashkari et al. [10] | Time-related features | Random Forest; C4.5; KNN | Self-collected, called Tor-nonTor (ISCXTor2016) | Precision; recall | Hodo et al. [11] | Time-related features using correlation-based feature selection | Artificial neural network; support vector machine | The same as [8] | Accuracy; precision; false positive rate | Almubayed et al. [12] | Features generated by NetMate (http://f00l.de/netmate/) | Naïve Bayes; C4.5; Random Forest; support vector machine | Self-collected | Precision; FP rate | Mayank and Singh [13] | Statistics calculated by NetAI (http://caia.swin.edu.au/urp/dstc/netai) | Random Forest; J4.8; AdaBoost | Self-collected | TP rate; FP rate; ROC Area | Cuzzocrea et al. [14] | Features calculated by ISCXFlowMeter (the tools implemented by [8]) | J4.8; BayesNet; jRip; OneR; RepTREE | Self-collected | TP rate; FP rate; precision; recall; F-measure; MCC; ROC Area; PRC area | Rao et al. [16] | Packet level and flow level features | Gravitational clustering | Self-collected | Rand statistic; Jaccard coefficient; FM; averaged accuracy | He et al. [16] | TLS/SSL-related features; packet size-related features | TLS fingerprint-based method; packet size distribution-based method | CAIDA Equinix Chicago (http://www.caida.org/data/passive/passive_2010_dataset.xml) | TP rate; false positive rate | Barker et al. [17] | TLS/SSL-related features | Just logical judgment | Self-collected | - | Bai et al. [18] | Traffic fingerprints; characteristic strings | AC-BM algorithm | Self-collected | Recognition rate; misrecognition rate | Zhioua [19] | Interpacket times | Hidden Markov models | Self-collected | Precision; F-measure |
|
|